Re: Indo-European Languages and Gramatical Gender Loss

On 16 Jun, 12:09, LEE Sau Dan <dan...@xxxxxxxxxxxxxxxxxxxxxxxxxx>
wrote:
"Seán" == Seán O'Leathlóbhair <jwlaw...@xxxxxxxxx> writes:

One last attempt to get my poinit across. I notice that no one else
has queried this point.

Seán> I am getting tired of this. I am quite familiar with
Seán> statistics, probability, distributions etc but I still don't
Seán> see that they are important here. I am questioning the
Seán> utility of gender in French.

As you said, the utility of gender is zero if words clash in gender.
So, when considering the utility, you have to take into account _how
often_ it is useful. That involves statistics, which is built upon
propbability theories, which in turn takes distribution as a
parameter.

You want more detail but every time I quote some numbers you snip
them. Here are some numbers again.

Distribution Probability of different gender

50/50 0.500
60/40 0.480
70/30 0.420
80/20 0.320

This shows that the distribution needs to deviate quite far from 50/50
before the probability drops significantly. Note that the 60/40
probability is still 0.48. I would be very surprised if the
distribution of gender in French was that far off 50/50.

Seán> If I can do that with an assumption of 50/50 then there is
Seán> no need to go into these complications.

That's to simplistic and unrealistic.

No, look at those figures again. If the true distribution is 70/30
then the probability of mixed gender is 0.42. By aiming slightly
higher and doubting the utility at 0.5, I don't need to care about the
distribution.

Seán> If you want to attack me in this area, I will help you. A
Seán> speaker may be able to increase the utility by choosing
Seán> synonyms which maximised the contrast.

Are such choices always available? If not, how often?

Well, I am giving you ideas how to criticise me here so I don't feel
the need to do all the work for you. Are suitable synonyms always
available? I very much doubt it. Are they never available? I also
doubt that. When these synonyms exist, does anyone select them to
achieve a gender contrast? I have no idea. At a guess, I would
expect that it sometimes happens in poetry and other carefully
constructed utterances but rarely in normal speech.

>> Checking a dictionary is a bad idea, as it is a biased sample.
>> By doing so, you're giving each word an equal chance. It
>> should not be difficult to realize that in most languages 20%
>> of the words are used 80% of the time. Your method doesn't
>> reflect the frequency of occurrence of each word. It's a
>> highly skewed sample.

Seán> Checking a dictionary is an easy way to get a rough
Seán> estimate.

A rough AND WRONG one. Because you're counting rare words with equal
weight as frequent word. The estimate would be far away from reality.

Seán> It may help you decide if it is worth the effort to gather
Seán> more data.

That's not an excuse to use a biased sample to do estimations.

If you can gather perfect data before any investigation or decision
then that's great. Over in the real world, it is not always
practical, possible, or economic to gather perfect data. Estimates
and imperfect data are often the only ones available. Or these
estimates or imperfect data need to be used to justify the cost of
gathering better data.

Seán> I can grasp the basics of statistics. Can you grasp that
Seán> they are not relevant in this case?

I can't. You're talking about how useful something is. How can you
ignore how often it can be applied?


One last attempt. I will use an analogy from maths. Suppose you want
to determine whether f(x) < g(x) for all x. This may be easy or hard
depending on the functions. Here's one example, f(x) = 1 - x^2 and
g(x) = x^2 -6x + 11. Some algebra or calculus shows that f(x) has a
maximum of 1 (at x = 0) and g(x) has a minimum of 2 (at x = 3). Since
the maxiumum of f(x) is less than the minimum of g(x) the job is
done. It is not necessary to check specific values of x. This is
comparble to my claim about the gender in French. I have calculated
the maximum probability of a gender distinction and claim that even at
that probability, it has little utility. If so, there is no need to
check specific values of x (distributions).

Of course, it is not always so easy. If we change g(x) to x^2 - 6x +
9 then it is still true that f(x) < g(x) for all x but the maximum of
f v minimum of g argument no longer works. g(x) has a minimum of 0 at
x=3. The minimum / maximum test does not prove f(x) < g(x) but
neither does it disprove it. More work is needed in this case.

So, the technqiue does not always work but it does sometimes work and
it is a useful tool to have available.

--
Seán O'Leathlóbhair

.

Relevant Pages

  • Re: Statistics Question
    ... I have reading a basic statistics text, ... probability I would get from a binomial distribution. ... Lets say I flip a coin 40 times, and get 16 heads. ... You are trying to estimate the *discrete* binomial distribution ...
    (rec.sport.baseball)
  • Re: Indo-European Languages and Gramatical Gender Loss
    ... I am questioning the utility of gender in French. ... and check these issues to see if the distribution was sufficiently ... Good statistics of actual frequencies of the two ... predict the probability of a gender contrast if there was no gender ...
    (sci.lang)
  • Re: Uniform probability on R and N?
    ... >an "Uniform probability distribution on the real line R"? ... properties of finite integers or reals. ... Herman Rubin, Department of Statistics, Purdue University ...
    (sci.stat.math)
  • Re: Incomplete knowledge and statistics
    ... >''statistical mechanics may be applied to a system ... It slightly depends on what you mean by statistics. ... >statistics since you know nothing at all about the distribution ... >of description, is a probability distribution, and that this probability ...
    (sci.physics.research)
  • Re: How do I claim a measurement is Gaussian distribution?
    ... I am a student from engineering. ... I have never taken a statistics ... but I do have background of probability. ... There was a recent thread here "Goodness of fitting of a distribution" ...
    (sci.stat.math)
Quantcast